Echium L. (Boraginaceae) (Speciation/Selection/Perennial Growth/Woodiness/Founding Populations) UTA-REGINA B6HLE*, HARTMUT H

Home , Brunnera macrophylla, Cerinthe, Echium, Echium creticum, Lithodora, Pulmonaria officinalis

Proc. Natl. Acad. Sci. USA Vol. 93, pp. 11740-11745, October 1996 Evolution

Island colonization and evolution of the insular woody habit in Echium L. (Boraginaceae) (speciation/selection/perennial growth/woodiness/founding populations) UTA-REGINA B6HLE*, HARTMUT H. HILGER*, AND WILLIAM F. MARTINtt *Institut fur Systematische Botanik und Pflanzengeographie, Freie Universitat Berlin, Altensteinstrasse 6, D-14195 Berlin-Dahlem, Germany; and tInstitut fur Genetik, Technische Universitat Braunschweig, Spielmannstrasse 7, D-38023 Braunschweig, Germany Communicated by Peter H. Raven, Missouri Botanical Garden, St. Louis, MO, August 1, 1996 (received for review March 19, 1996)

ABSTRACT Numerous island-inhabiting species of pre- mism is strict; no island species occur naturally in continental dominantly herbaceous angiosperm genera are woody shrubs floras and vice versa. Second, insular woodiness is very or trees. Such "insular woodiness" is strongly manifested in strongly pronounced in the genus; all but two of the island the genus Echium, in which the continental species of circum- species (Echium bonnetii and Echium pitardii) are woody mediterranean distribution are herbaceous, whereas endemic perennials, whereas the continental species are annual to species of islands along the Atlantic coast of north Africa are perennial herbs (or very rarely subshrubs). Third, the ages of woody perennial shrubs. The history of37 Echium species was the islands on which the woody species occur are known traced with 70 kb of noncoding DNA determined from both (9-14), so that the evolutionary process can be correlated to chloroplast and nuclear genomes. In all, 239 polymorphic geologic time. Island species of Echium reveal a degree of positions with 137 informative sites, in addition to 27 infor- morphological diversity comparable to that found in the mative indels, were found. Island-dwelling Echium species are remaining ("2000) species of the family. This is reflected in shown to descend from herbaceous continental ancestors via woody habits of the perennial island species: candelabra shrubs a single island colonization event that occurred <20 million and monocarpic rosette trees, the large vegetative rosettes of years ago. Founding colonization appears to have taken place which are formed over several vegetation periods and produce on the Canary Islands, from which the Madeira and Cape single inflorescences of several thousand flowers prior to Verde archipelagos were invaded. Colonization of island hab- fruiting and death. The herbaceous continental species reveal itats correlates with a recent origin of perennial woodiness much less diversity in form and preferentially inbreed, whereas from herbaceous habit and was furthermore accompanied by the island species are preferential outbreeders (15, 16). The intense speciation, which brought forth remarkable diversity Macaronesian species inhabit virtually all characteristic vegeta- offorms among contemporary island endemics. We argue that tion zones of the islands: coastal rocks, cliffs, semidesert habitats, the origin of insular woodiness involved response to counter- the laurel forest region, and the volcanic subalpine zone. selection of inbreeding depression in founding island colonies. Wishing to better understand the evolution of perennial woody habits in Echium and the phylogenetic process behind That island-dwelling species of numerous, predominantly her- diversity of its island species, we have inferred the phylogeny baceous genera of flowering plants tend to exhibit a perennial of 37 Echium species-including most of the island inhabit- and woody growth form was evident to Darwin (1), Wallace ants-and four outgroup representatives by amplifying, se- (2), and many others since. Carlquist (3) coined the term quencing, and analyzing noncoding DNA regions from both "insular woodiness" to designate the phenomenon (which also their chloroplast and nuclear genomes. The data reveal that is observed in highland biota; refs. 4 and 5) and found evidence perennial, woody Macaronesian Echium species arose <20 that wood morphology and development in such plants differ million years ago (mya) from herbaceous continental anteced- from that of common woody tissues. The insular woody habit ants and that all contemporary island dwellers can be traced to occurs in several angiosperm families, among them Boragi- a single founding colonization event emanating from conti- naceae, Asteraceae, Campanulaceae, Lobeliaceae, and Eu- nental forms. Evidence that the origin of insular woodiness in phorbiaceae, and was suggested to have evolved several times Echium was crucial to colonization-related diversification of independently from the herbaceous as an adaptation to cli- island forms supports and extends Wallace's suggestion (2) matic (primarily release from seasonality) and other factors that reproduction rather than adaptation is ultimately causal to (e.g., absence of large herbivores; ref. 3). According to the perennial woody habit in island species. opposing relict hypothesis, which draws support from morphological, cytological, and paleogeographical data, the woody habit of island forms is considered ancestral to the herbaceous MATERIALS AND METHODS habit (6-8). Whereas polarity of evolution for perennial Twenty-four of the 27 island Echium species, in addition to 13 woodiness in island species has been heavily debated (2-8), continental species predominantly of west Mediterranean problems concerning the underlying evolutionary mechanisms flora and four confamilial outgroup representatives, were through which it arose have remained largely intractable. investigated (Table 1). The outgroup representatives investi- An excellent case example of insular woodiness (Fig. 1) is gated were L. fruticosus Buek, Cerinthe minor L., Cerinthe found in Echium L. (Boraginaceae). Its size and distribution major L., and Lithodora diffusa (Lag.) I. M. Johnston. Material are particularly well-suited to study the phenomenon in several was collected either in the field or from various botanical respects. First, 27 species of Echium are endemic to the gardens and herbariums. All details of sources for all individ- Canary, Madeira, and Cape Verde archipelagos of the Mac- ual accessions are available from the authors ronesian islanids, whereas the remaining ("30) congeners are upon request. found in circummediterranean-west Asian floras. Their ende- Abbreviations: cpDNA, chloroplast DNA; mya, million years ago. Data deposition: The sequences reported in this paper have been The publication costs of tlfis lircle were defrayed in part by page charge deposited in the GenBank data base (accession nos. L43165-L43324). payment. This article must therefore be hereby marked "advertisement" in ITo whom reprint requests should be addressed. e-mail: w.martin@ accordance with 18 U.S.C. §1734 solely to indicate this fact. tu-bs.de. 11740 Downloaded by guest on September 29, 2021 EVOlUtiOn: BOh1e et al. Proc. Natl. Acad. Sci. USA 93 (1996) 11741

trnLUAA-trnFGAA Spacer Continental species ... TTACACAAGTCTTGTATGATA... Island species ...TTACA------TATGATA...

trnLUAA Intron Continental species ...GACT-----GCTTTGG... except E. sabulicola ... GCCTGCTTTGCTTTGG... E. parviflorum ... GCCTGCTTTGCTTTGG... Island species ... GACTGCTTTGCTTTGG... FIG. 2. Phylogenetic distribution of two indels found in Echium DNA sequences. informative indels (alignment available from the authors upon request). The trnLuAA intron shows least variability among all regions investigated (Table 2); average divergence of the nuclear ITS1 is 4- to 20-fold higher than that of the cpDNA spacers, depending on the species group under consideration (Table 2). Nuclear divergence within the island species is remarkably lower than within their continental congeners, ranging from 0.02-0.2% to 1.9% for cpDNA markers and ITS1, respectively (Table 2). We found several length polymorphisms (indels) in the chloroplast data set, two of which are either unique to the island species investigated or to the latter and two of the continental species, E. parviflorum and E. sabulicola (Fig. 2). Neighbor-joining and parsimony analyses of the combined FIG. 1. Echium creticum (Left) and Echium simplex (Right), representatives of the herbaceous continental and woody island species, data set of all four markers (Fig. 3) are totally consistent with respectively, found in this genus of circummediterranean and Maca- the indel distribution, providing unequivocal support for the ronesian (Canary, Madeira, and Cape Verde archipelagos) distribu- common descent of all island species investigated. Within the tion. E. creticum is annual to biennial, whereas E. simplex, an endemic continental species, the two coastal plants E. parviflorum and of the Anaga mountains (Tenerife), is a monocarpic rosette tree-i.e., E. sabulicola are more closely related to the Macaronesian forms a perennial vegetative rosette and flowers only once in life, species than to continental relatives, and represent the sister producing several thousand propagules and dies. group to the former. L. fruticosus was originally studied as an outgroup representative, since morphological evidence had DNA was isolated from leaf material (fresh, dried, or pre- suggested that this South African genus is closely related to served in 70% ethanol) as described (21), extracted with Echium. Yet L. fruticosus and all other members of that genus phenol/chloroform, and purified by ultrafiltration in Microcon investigated so far proved to be ingroups (unpublished data) 100 devices (Amicon). Amplification was performed in 10 mM due to the unexpectedly deep outlying position of the west Tris-HCl (pH 8.3), 50 mM KCl, 2 mM MgCl2, 1 mM of each Asian herb E. russicum (Fig. 3); the genus Echium as previ- dNTP, and 1 ,kM of each primer containing 0.02 units/,ul Taq ously defined is paraphyletic and therefore under revision polymerase (Perkin-Elmer) for 35 cycles of 1 min at 93°C, 1 (unpublished data). Trees constructed from each of the four min at 50°C, and 2 min at 72°C. An aliquot of the double- markers individually were generally consistent with the topol- stranded product was used as template for asymmetric PCR in ogy in Fig. 3, yet provided low bootstrap support for internal which one of the primers was diluted 100-fold. Single-stranded branches due to the small number of sites sampled in each templates were purified by ultrafiltration (as above) and (data not shown). Neighbor-joining and parsimony trees con- sequenced as described (22). Primers used for amplification structed from the combined cpDNA data (61 informative sites, and sequencing of chloroplast trnTuGu-trnLUAA spacer, trn- 22 informative indels) were completely congruent at the 80% LUAA intron, and trnLUAA-trnFGAA spacer were those de- bootstrap proportion interval both with nuclear ITS1 trees scribed in ref. 23; additional internal sequencing primers were (data not shown) and with Fig. 3. Congruence between 5'-GAATAGCTAATTTCAAATAA-3' and 5' -ATAG- supported regions of the parsimony and neighbor-joining GAGACTCTTATGTTGG-3'. Primers used for amplification topologies is consistent with the finding that, in computer and sequencing of the nuclear 18S-5.8S ITS1 region of rDNA simulations under the given sequence parameters for our data were taken from the literature (24), and an additional internal (reasonably large number of nucleotides sampled per species sequencing primer, 5'-GTCGTAACAAGGTTTCTGTAG- and low overall sequence divergence), both methods perform 3', was used. Sequences were aligned visually; phylogenetic efficiently (29, 30). A maximum likelihood tree constructed analyses were performed with maximum parsimony (DNAPARS with DNAML of PHYLIP also found the same topology as shown of PHYLIP version 3.53c; ref. 25) and neighbor-joining (26) in Fig. 3, with the exception of differences in branching order using the Kimura distance (27). All sequences reported here within island species, and all branches supported at the 80% have been deposited in the GenBank data base under the bootstrap proportion level in Fig. 3 were significantly positive accession nos. L43165-L43324. in the maximum likelihood tree at the P = 0.01 level. Using Lobostemon as the outgroup, the relative rate test was per- RESULTS formed as described (31) for concatenated nuclear and chloroplast sequences; no difference in substitution rate for any Minimum and maximum sequence lengths observed for the pair of ingroup Echium species was found that was significant three chloroplast DNA (cpDNA) regions within Echium are at the P = 0.05 level. 611/684 bp (trnTuGu-trnLUAA spacer), 477/488 bp (trnLUAA intron), and 349/358 bp (trnLuAA-trnFGAA spacer); the length of the nuclear 18S-5.8S ITS1 region varies from 230 to 260 bp. DISCUSSION The alignment of all four markers contains 239 polymorphic One Mainland-to-Island Invasion Followed by Speciation. positions, 137 phylogenetically informative positions, and 27 Phylogeny correlates with biogeography in Echium, as shown Downloaded by guest on September 29, 2021 11742 Evolution: B6hle et al. Proc. Natl. Acad. Sci. USA 93 (1996)

Table 1. Geographic distribution and habit of investigated species Species Habit Distribution Echium albicans Lag. & Rodr. Annual-perennial herb* S Spain E. asperrimum Lam. Biannual herb W-med E. creticum L.t Annual-biannual herb W-med E. horridum Batt. Annual herb SW-med E. italicum L. Biannual herb NNE-med E. lusitanicum L. Perennial herb Iber E. parviflorum Moench. Annual-biannual herb Circummed E. plantagineum L.t Annual herb Circummed§ E. rosulatum LangeJ Biannual-perennial herb* Iber E. russicum Gmel. Biannual herb pont-pan E. sabulicola Pomel Annual-perennial herb* W-med E. tuberculatum Hoffm. & Linkll Biannual herb W-med E. vulgare L. Biannual herb Eu, W Asia E. aculeatum Poiret Candelabra tree Can-C, -G, -H, -T E. auberianum Webb & Berth. Monocarpic herb** Can-T E. brevirame Sprague & Hutch. Candelabra tree Can-P E. bonnetii Coincy Annual herb Can-C, -F, -T E. callithyrsum Webb ex Bolle Candelabra tree Can-C E. decaisnai Webb & Berth. Candelabra tree Can-C, -F, -L E. giganteum L.f. Candelabra tree Can-T E. handiense Svent. Candelabra tree Can-F E. hierrense Webb ex Bolle Candelabra tree Can-H E. leucophaeum Webb Candelabra tree Can-T E. onosmifolium Webb & Berth. Candelabra tree Can-C E. pininana Webb & Berth. Monocarpic rosette-tree Can-P E. pitardii A. Chev. Annual herb Can-L E. simplex DC. Monocarpic rosette-tree Can-T E. strictum L.f. Candelabra tree Can-C, -G, -H, -P, -T E. triste Svent. Annual-perennial herb* Can-C, -G, -T E. virescens DC. Candelabra tree Can-T E. webbii Coincy Candelabra tree Can-P E. wildpretii Pears. Monocarpic rosette-tree Can-P, -T E. hypertropicum Webb Candelabra tree Cape V-A, -F, -R, -T E. stenosiphon Webb Candelabra tree Cape V-A, -L, -N, -V E. vulcanorum A. Chev. Candelabra tree Cape V-F E. candicans L.f. Candelabra tree Mad-M, -Po E. nervosum Ait. Candelabra tree Mad-M, -Po, -D Lobostemon fruticosus Buek Shrub S Africa Distribution data were taken from the literature (17-20). S, south; W, west; SW, southwest; NNE, north northeast; med, Mediterranean; Circummed, circummediterranean; Iber, Iberian peninsula; pont-pan, pontic-pannonic; Eu, Europe; Can-C, -F, -G, -H, -L, -P, and -T, Canary Islands-Gran Canoria, -Fuerteventura, -Gomera, -Hierro, -Lanzarote, -La Palma, and -Tenerife, respectively; Cape V-A, -F, -L, -N, -R, -T, and -V, Cape Verde archipelago-Santo Antao, -Fogo, -Santa Luzia, -Sao Nicolau, -Brava, -Sao Tiago, and -Sao Vincente, respectively; and Mad-M, -Po, and -D, Madeira Islands-Madeira, -Porto Santo, and -Desertas, respectively. *Rarely subshrubs. In the last monograph of the genus (18), species names are given as tE. spinescens Medikus, tE. lycopsis Grufb., SE. gaditanum Boiss., and IE. hoffmannseggii Litard. §F. plantagineum occurs worldwide through human distribution. **E. auberianum rarely buds again after flowering and is thus not strictly monocarpic. in Fig. 3. The congruence that we observed between the these may be older than and ancestral to their morphologically cpDNA (maternally inherited in Echium; ref. 32) and nuclear less diverse herbaceous continental congeners, but the molec- DNA trees indicate that pollen introgression from continental ular data clearly indicate that the converse is true (Fig. 3). to island species (or vice versa) has not played a detectable role Since the island species of Echium are obviously successful in Echium evolution, a point which cannot be established long-range colonizers, it is curious that no continental Echium through analysis of cpDNA alone. The data provide convinc- species studied is a descendant of the island inhabitants-i.e., ing support for the view that island species of Echium are no fruitful reinvasion of the mainland has occurred within our derived descendants of continental ancestors. The molecular sample. phylogeny based on four markers from two compartments is The Timing of Island Colonization. The age of the oldest consistent with the indel distribution (Fig. 2), indicating a islands among the three archipelagos are known and provide single mainland-to-island invasion event in the history of this the earliest possible dates of island colonization. Within the genus followed by accelerated speciation among island colo- Canary Islands, eastern-most Fuerteventura and Lanzarote nizers, rather than multiple invasions (18). Notably, the Ca- are the oldest and subaeric formation started -20 mya and nary, Madeira, and Cape Verde archipelagos, though sepa- 15 mya on these islands, respectively, as indicated by K/Ar rated by several thousand kilometers of ocean, have all been determinations (9), whereby westernmost Hierro is the young- colonized by descendants of the same initial invading popu- est (-0.8 mya; ref. 10). A similar east-to-west gradient of lation. Under the relict hypothesis (6), the conspicuous diver- Miocene to Quaternary age is found for the Cape Verde sity among Macaronesian Echium species would suggest that archipelago (11-13), and the younger Madeira archipelago is Downloaded by guest on September 29, 2021 Evolution: B6hle et al. Proc. Natl. Acad. Sci. USA 93 (1996) 11743

Table 2. Average divergence between taxa for sequences studied colonization, possibly reflecting a nascent reiteration of the Mini- Maxi- initial mainland-to-island colonization/diversification process. Comparison (n) Marker Kav SE mum mum Colonization of Madeira from the Canary Islands would involve dispersal against the northeast-to-southwest current of Outgroups/Continental (36) nITS* 303 72 227 372 both prevailing winds and ocean currents surrounding all three trnT/Lt 34 13 18 49 archipelagos and would be compatible with the view that birds trnL/F* 29 14 16 45 are responsible for long-distance seed dispersal in island Intron§ 28 12 15 39 species of Echium. Outgroups/Island (63) nITS 293 70 243 328 Evolution from Herbaceous to Radiating Woody Species. trnT/L 25 11 15 31 Outgroup comparison to the closest continental sisters (E. trnL/F 31 16 23 39 parviflorum and E. sabulicola) of woody Echium island species Intron 26 11 13 34 strongly suggests that the ancestors of the latter were herba- Continental/Island (252) nITS 97 33 68 119 ceous. Although the data do not thoroughly exclude the trn/L 19 10 6 32 alternative explanation that the herbaceous habit evolved from trnL/F 14 10 3 29 woody ancestors (6-8), that scenario appears rather unlikely, Intron 4 5 2 7 since it would have necessarily occurred independently in Continental/Continental (66) nITS 73 26 0 146 several different continental lineages (Fig. 3). Evolution of trnT/L 17 9 0 35 woody from herbaceous forms is by far the most straightfor- trnL/F 9 7 0 22 ward interpretation of the data, and is congruent with similar Intron 6 6 0 11 conclusions recently drawn from cpDNA restriction mapping Island/Island (210) nITS 19 12 0 54 for the genus Lobelia (Lobeliaceae; ref. 33). The position of the trn/L 2 3 0 8 herbaceous island inhabitants E. bonnetii and E. pitardii (which trnL/F 1 3 0 3 on the basis of ITS1 data alone is the immediate sister of E. Intron 0.2 0.4 0 2 bonnetii; data not shown) is not robustly resolved as basal or Cape Verde/Cape Verde (3) nITS 3 3 0 4 "crown" within the island phylogeny (Fig. 3). Although the Madeira/Madeira (1) nITS 0 0 0 0 data do not clearly reveal whether their herbaceous states are Values represent average Kimura two-parameter distance x 10-3 homologous to that of continental sisters (ancestral) or (Ka,) between outgroup representatives (C. minor L., C. major L., and whether it represents a recent reversal from the insular woody Lithodora diffusa Johnston, each Boraginaceae), continental (except habit, the position of E. bonnetii in the tree would favor the E. russicum), island, Cape Verde, and Madeira Echium species, view that it has undergone very recent reversal since its respectively. SE, average standard error of Kimura distance across divergence from a woody island ancestor. comparisons. n, number of individual pairwise comparisons. Mini- very low of observed be- mum and maximum values for taxon comparisons are indicated. The degree genetic divergence *Nuclear ITS1 (average, 235 sites). tween island Echium species (Table 2) combined with remark- tChloroplast trnTuGu-trnLuAA spacer (average, 720 sites). able morphological diversity (Fig. 3) clearly indicates that they tChloroplast trnLuAA-4rnFGAA spacer (average, 312 sites). are still undergoing active diversification and that speciation §Chloroplast trnLuAA intron (average 493 sites). occurs directly in conjunction with island colonization. This is congruent with conclusions recently drawn for other island of mid-Miocene age (10-15 mya; ref. 14). Since the Macaro- genera on the basis of less comprehensive data (34, 35). Note nesian species share a single common ancestor, Echium island that a prominent speciation mechanism in angiosperms, ploidy inhabitants and corresponding speciation events thus cannot change (36, 37), can be excluded for island inhabitants of be >20 million years old (9-14) and could be much younger. Echium, since they have constant chromosome number (2n = Using this upper bound, the substitution rate for nuclear ITS 16; refs. 38 and 39), whereas their continental sisters show sequences within island species ofEchium can be estimated as considerable ploidy variation with deviations from x = 8 (n = -1.1 x 10-9 per site per year (at the minimum) on the basis 8, n = 16, x = 6, x = 5, x = 7; refs. 40 and 41). E. parviflorum of average divergence (Table 2) between the three earliest and E. sabulicola are clearly close continental relatives of the branching (E. auberianum, E. pininana, and E. wildpretii) and island species (Figs. 2 and 3) and also possess 2n = 16 (41), the remaining island species (63 comparisons, Kav = 0.042 + further supporting their position as the sister group to the 0.02). This value is =4-fold lower than that found in vertebrate island colonizers. Speciation of the older and cytologically nuclei (31), but, conversely, if nuclear substitution rates in more polymorphic continental Echium forms might thus have Echium and vertebrates are, in fact, similar, the island species entailed chromosome evolution, but rapid speciation among would be much younger (-5 million years old) than the islands the woody island Echium species clearly did not. which they inhabit. Using the conservative rate estimate of We have found that in Echium (i) only one initial invasion 1.1 x 10-9 per site per year, divergence between nuclear ITS of the islands from the continent has occurred, (ii) the sequences indicates that the three Echium species endemic to perennial woody forms are very proficient colonizers, and (iii) the Cape Verde archipelago, E. hypertropicum and E. vulca- speciation occurs in conjunction with island colonization (even norum shown in Fig. 3 (and E. stenosiphon which differs from across ocean distances much greater than that to the conti- these in only one ITS substitution, not shown), diverged not nent). Taken together, these findings compellingly suggest that more than 3 mya and perhaps as recently as 0.75 mya. From the critical and rate-limiting step in the evolutionary success Table 2, the same ages can be estimated for the (only) two (measured in terms of both new species and forms per unit Echium endemics of the Madeira archipelago (E. candicans, E. time) of the contemporary island species was the origin of the nervosum). perennial woody habit. Insular woodiness is crucial to under- The greatest genetic divergence within the island species is standing colonization-related diversity found in the genus. found among inhabitants of the Canary Islands, suggesting that Possible Causes of Insular Woodiness. That insular wood- the initial colonization and diversification occurred on this iness as well as candelabra shrub and rosette tree habits have archipelago; the Madeira and Cape Verde archipelagos were arisen independently in many different angiosperm genera (3) apparently colonized later by perennial woody Canary Island strongly suggests that they are selected for, rather than occur- inhabitants (Fig. 3 and Table 2). Common branches for ring as the result of drift. But what type(s) of selection can Madeira and Cape Verde inhabitants, respectively, suggest account for the woody habit of island colonizers? If selection that Echium speciation on these islands-like the initial inva- had entailed adaptation to constant climates as suggested (3), sion-again has occurred subsequent to a single founding one would expect Echium inhabitants of comparable (i.e., Downloaded by guest on September 29, 2021 11744 Evolution: B6hle et aL Proc. Natl. Acad. Sci. USA 93 (1996)

FIG. 3. Correlation between phylogeny and biogeography in Echium. On the left side of the figure is a neighbor-joining tree (26) for Kimura distances (27) between combined sequences from noncoding cpDNA and nuclear DNA. The horizontal scale bar indicates 1% sequence divergence. The tree is based on a total of 1676-1790 bases per species, and the alignment contains 239 polymorphic positions, 137 informative positions, and 27 informative indels (available from the authors upon request). Numbers above branches indicate the number of times the branch was found in 100 bootstrap neighbor-joining replicates (28) using the Kimura distance, numbers below branches indicate the number of times the branch was found in 100 bootstrap parsimony replicates. Bootstrap values <80 are not indicated. H, herbaceous growth form; W, woody growth form. In the middle portion of the figure is a schematic representation of representative habits. Vertical scale bars indicate 10 cm. In the right portion of the figure is geographic distribution of species (the lighter shaded region indicates the distribution of E. russicum). Outgroup representatives are C. minor L., C. major L., and Lithodora diffusa Johnston (each Boraginaceae). Use ofthe unpublished sequences for these four regions from Symphytum officinale L., Pulmonaria officinalis L., Anchusa officinalis L., and Brunnera macrophylla (Adams) I. M. Johnston (all Boraginaceae) as outgroups changes neither the position of the root nor regions of the topology supported at the 80% confidence interval. The island species shown do not occur in natural mainland floras, nor do continental species naturally occur on Macaronesian islands (indicated by bold lines). Circummediterranean grey shading indicates the cumulative distribution of continental Echium species; distribution of E. russicum and Lobostemon, respectively, are indicated. Schematic drawings of Echium species are redrawn from ref. 18. (1) Echium plantagineum; (2) E. humile; (3) E. rosulatum; (4) E. parviflorum; (5) E. onosmifolium; (6) E. decaisnai; (7) E. simplex; (8) E. wildpretii; and (9) Lobostemon regulareflorus. Sequences of E. stenosiphon, E. triste, and E. pitardii of the Cape Verde and Canary archipelagos, respectively, were not obtained for the trnTUGu-rnLUAA spacer and are thus not included in the figure, but these species are descendants of the same invasion as documented by sequence data from the remaining markers.

Iberian) continental climate zones also to be woody, but they But in light of our surprising findings that the island forms are herbaceous. In our view, this is a strong argument against are rapidly speciating colonizers, their outbreeding behavior the climate adaptation hypothesis. Wallace's earlier suggestion seems puzzling at first sight, since, in an extreme scenario, one that a connection may exist between the perennial woody habit could expect successful island colonizers to be able to establish of island species and reproductive strategy deserves closer populations with but a single propagule (42). Although selec- inspection. He argued that the perennial insular woody growth tion for successful pollination (2) can fully account for the form primarily reflects selection for longevity (rather than for woodiness (longevity) and large inflorescences observed in woodiness per se) of insect-pollinated species in an environ- island Echium species, it does not explain the requirement for ment where insects initially should be expected to be rare (2), cross-pollination in island inhabitants. We suggest that out- noting that the resulting increase in size may have provided breeding is manifest among island inhabitants as the result of additional advantage in niche competition among initial col- counterselection of inbreeding depression in selfing coloniz- onizers. Bramwell's studies of breeding behavior are generally ers. This view derives strong support from the finding that compatible with that view, since in Echium species studied, the natural per generation mutation rates in inbreeding herba- woody island forms were found to be outbreeding, setting only ceous members of the Boraginaceae are sufficiently high to 0-11% fertile seed in selfings, whereas their continental sisters cause inbreeding depression in newly arisen populations (43). as well as the herbaceous island inhabitant E. bonnetii were If outbreeding is the primary selective factor in island colo- preferentially inbreeding (15, 16). nization, pollination pressure will subsequently favor rare, Downloaded by guest on September 29, 2021 Evolution: Bohle et al. Proc. Natl. Acad. Sci. USA 93 (1996) 11745

large, conspicuous inflorescences among outbreeders and, as 13. Stillman, C. J., Furnes, H., LeBas, M. J., Robertson, A. H. F. & a consequence, select perennial (and therefore woody) habits Zielonka, J. (1982) J. Geol. Soc. (London) 139, 347-361. capable of producing them, in agreement with Wallace's 14. Zbyszewski, G. (1980) Congr. Geol. Int. 26, 108-114. salient arguments (2). Under this view, diversity of contem- 15. Bramwell, D. (1973) Monogr. Biol. 4, 71-82. porary woody Echium forms reflects a multiplicity of select- 16. Bramwell, D. (1972) Int. Organ. Plant Biosyst. Newsl. 6, 2-9. able developmental pathways toward longevity, rather than 17. Hansen, A. & Sunding, P. (1993) Sommerfeltia 17, 1-295. selection for specifically 18. Klotz, G. (1959) Die Systematische Gliederung der Gattung environment-adapted variants of such Echium L., ein Beitrag zum Problem der Gliederung bei Pflanzen, woody perennial habits as schematically depicted in Fig. 3. In Habilitationsschrift (Martin-Luther-Universitat Halle, Germa- other words, insular woodiness in Echium might simply betray ny). "survival of the founders," and many differences between 19. Levyns, M. R. (1934) J. Linn. Soc. London Bot. 49, 393-451. perennial woody habits could be nonadaptive. The proposal 20. Bramwell, D. & Bramwell, Z. (1983) Flores Silvestres de las Islas that avoidance of inbreeding depression may be causal to the Canarias (Rueda, Madrid). perennial woody growth form of island species can easily 21. Doyle, J. J. & Doyle, J. L. (1990) Focus (Rochester, N.YY) 12, account for available data in Echium and also for the finding 13-15. that gynodioecy, a common strategy to achieve outbreeding, 22. Sanger, F., Nicklen, S. & Coulson, A. R. (1977) Proc. Natl. Acad. although observed in several island Echium populations, does Sci. USA 74, 5463-5467. not correlate with perennial woodiness, since it is also found 23. Taberlet, P., Gielly, L., Patou, G. & Bouvet, J. (1991) Plant Mol. in herbaceous continental Echium species (16, 44-46). It Biol. 17, 1105-1109. furthermore predicts that estimates of outcrossing rates (47, 24. Bobola, M. S., Smith, D. E. & Klein, A. S. (1992) Mol. Biol. Evol. 48) in natural populations of insular woody Echium species 9, 125-137. should be than in those of 25. Felsenstein, J. (1993) PHYLIP, Phylogeny Inference Package higher herbaceous continental (Univ. of Washington, Seattle), Version 3.5c. congeners. Outbreeding, perennial growth and the insular 26. Saitou, N. & Nei, M. (1987) Mol. Biol. Evol. 4, 406-425. woody habit may reflect three manifestations attributable to a 27. Kimura, M. (1980) J. Mol. Evol. 16, 111-120. single selective pressure: the need to escape inbreeding de- 28. Felsenstein, J. (1989) Cladistics 5, 164-166. pression in geographically isolated founding populations. 29. Jin, L. & Nei, M. (1990) Mol. Biol. Evol. 7, 82-102. 30. Nei, M. (1996) Annu. Rev. Genet. 30, in press. We thank the Gesellschaft fur Biotechnologische Forschung for the 31. Li, W.-H. & Tanimura, M. (1987) Nature (London) 326, 93-96. generous use of their computer facilities, Jens Pahnke for stimulating 32. Corriveau, J. L. & Coleman, A. W. (1988) Am. J. Bot. 75, discussions, and David Bramwell and Julia Perez de Paz for numerous 1443-1458. critical comments on the manuscript. This work was funded by grants 33. Knox, E. B., Downie, S. R. & Palmer, J. D. (1993) Mol. Biol. Evol. from the Deutsche Forschungsgemeinschaft (Hi 329/6). U.-R.B. was 10, 414-430. the recipient of stipends from the Land Berlin and the Deutsche 34. Valdebenito, H., Stuessy, T. F., Crawford, D. J. & Silva 0. M. Forschungsgemeinschaft. (1992) Syst. Bot. 17, 470-480. 35. Sang, T., Crawford, D., Kim, S.-C. & Stuessy, T. F. (1994) Am. J. 1. Darwin, C. (1860) On the Origin of Species (Murray, London). Bot. 81, 1494-1501. 2. Wallace, A. R. (1878) Tropical Nature and Other Essays (Mac- 36. Stebbins, G. L. (1971) Chromosomal Evolution in Higher Plants millan, London). (Arnold, London). 3. Carlquist, S. (1974) Island Biology (Columbia Univ. Press, New York). 37. Ehrendorfer, F. (1976) in Origin and Early Evolution ofAngio- 4. Knox, E. B. & Palmer, J. D. (1995) Proc. Natl. Acad. Sci. USA 92, sperms, ed. Beck, C. B. (Columbia Univ. Press, New York), pp. 10349-10353. 220-240. 5. Knox, E. B. & Palmer, J. D. (1995) Am. J. Bot. 82, 1567-1573. 38. Borgen, L. (1969) Nytt Mag. Bot. 16, 81-121. 6. Bramwell, D. (1972) in Taxonomy, Phytogeography and Evolution, 39. Borgen, L. (1970) Nytt Mag. Bot. 17, 145-161. ed. Valentine, D. H. (Academic, London), pp. 141-159. 40. Fritsch, B. (1973) Bot. Not. 126, 450-458. 7. Mabberley, D. J. (1974) New Phytol. 73, 977-984. 41. Luque, T. (1984) Lagascalia 13, 17-38. 8. Meusel, H. (1952) Flora (Jena) 139, 333-393. 42. Baker, H. G. (1955) Evolution 9, 347-349. 9. Coello, J., Cantagrel, J.-M., Hernan, F., Fiuster, J.-M., Ibarrola, 43. Johnston, M. 0. & Schoen, D. J. (1995) Science 267, 226-229. E., Ancochea, E., Casquet, C., Jamond, C., Diaz de Teran, J.-R. 44. de Coincy, A. (1903) Bull. Herb. Boiss. Ser. 2 4, 261-277. & Cendrero, A. (1992) J. Volcanol. Geotherm. Res. 53, 251-274. 45. de Coincy, A. (1903) Bull. Herb. Boiss. Ser. 2 4, 488-499. 10. Fuster, J.-M., Hernan, F., Cendrero, A., Coello, J., Cantagrel, 46. Darwin, C. (1877) The Different Forms ofFlowers on Plants ofthe J.-M., Ancochea, E. & Ibarrola, E. (1993) Bol. R. Soc. Esp. Hist. Same Species (Murray, London). Nat. Secc. Geol. 88, 85-97. 47. Fritsch, P. & Riesenberg, L. H. (1992) Nature (London) 359, 11. Rothe, P. (1982) Cour. Forschungsinst. Senckenberg 52, 1-9. 633-636. 12. Boekschoten, G. J. & Manuputty, J. A. (1993) Cour. Forschungs- 48. Jarne, P. & Charlesworth, D. (1993) Annu. Rev. Ecol. Syst. 24, inst. Senckenberg 159, 3-5. 441-466. Downloaded by guest on September 29, 2021